src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_avx_256_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_avx_256_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  76     real             *charge;
  77     int              nvdwtype;
  78     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  79     int              *vdwtype;
  80     real             *vdwparam;
  81     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  82     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  83     __m256d          dummy_mask,cutoff_mask;
  84     __m128           tmpmask0,tmpmask1;
  85     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  86     __m256d          one     = _mm256_set1_pd(1.0);
  87     __m256d          two     = _mm256_set1_pd(2.0);
  88     x                = xx[0];
  89     f                = ff[0];
  90
  91     nri              = nlist->nri;
  92     iinr             = nlist->iinr;
  93     jindex           = nlist->jindex;
  94     jjnr             = nlist->jjnr;
  95     shiftidx         = nlist->shift;
  96     gid              = nlist->gid;
  97     shiftvec         = fr->shift_vec[0];
  98     fshift           = fr->fshift[0];
  99     facel            = _mm256_set1_pd(fr->epsfac);
 100     charge           = mdatoms->chargeA;
 101     krf              = _mm256_set1_pd(fr->ic->k_rf);
 102     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 103     crf              = _mm256_set1_pd(fr->ic->c_rf);
 104     nvdwtype         = fr->ntype;
 105     vdwparam         = fr->nbfp;
 106     vdwtype          = mdatoms->typeA;
 107
 108     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 109     rcutoff_scalar   = fr->rcoulomb;
 110     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 111     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 112
 113     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 114     rvdw             = _mm256_set1_pd(fr->rvdw);
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = jnrC = jnrD = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120     j_coord_offsetC = 0;
 121     j_coord_offsetD = 0;
 122
 123     outeriter        = 0;
 124     inneriter        = 0;
 125
 126     for(iidx=0;iidx<4*DIM;iidx++)
 127     {
 128         scratch[iidx] = 0.0;
 129     }
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 147
 148         fix0             = _mm256_setzero_pd();
 149         fiy0             = _mm256_setzero_pd();
 150         fiz0             = _mm256_setzero_pd();
 151
 152         /* Load parameters for i particles */
 153         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 154         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm256_setzero_pd();
 158         vvdwsum          = _mm256_setzero_pd();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             jnrC             = jjnr[jidx+2];
 168             jnrD             = jjnr[jidx+3];
 169             j_coord_offsetA  = DIM*jnrA;
 170             j_coord_offsetB  = DIM*jnrB;
 171             j_coord_offsetC  = DIM*jnrC;
 172             j_coord_offsetD  = DIM*jnrD;
 173
 174             /* load j atom coordinates */
 175             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 176                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 177                                                  &jx0,&jy0,&jz0);
 178
 179             /* Calculate displacement vector */
 180             dx00             = _mm256_sub_pd(ix0,jx0);
 181             dy00             = _mm256_sub_pd(iy0,jy0);
 182             dz00             = _mm256_sub_pd(iz0,jz0);
 183
 184             /* Calculate squared distance and things based on it */
 185             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 186
 187             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 188
 189             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 190
 191             /* Load parameters for j particles */
 192             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 193                                                                  charge+jnrC+0,charge+jnrD+0);
 194             vdwjidx0A        = 2*vdwtype[jnrA+0];
 195             vdwjidx0B        = 2*vdwtype[jnrB+0];
 196             vdwjidx0C        = 2*vdwtype[jnrC+0];
 197             vdwjidx0D        = 2*vdwtype[jnrD+0];
 198
 199             /**************************
 200              * CALCULATE INTERACTIONS *
 201              **************************/
 202
 203             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 204             {
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             qq00             = _mm256_mul_pd(iq0,jq0);
 208             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 209                                             vdwioffsetptr0+vdwjidx0B,
 210                                             vdwioffsetptr0+vdwjidx0C,
 211                                             vdwioffsetptr0+vdwjidx0D,
 212                                             &c6_00,&c12_00);
 213
 214             /* REACTION-FIELD ELECTROSTATICS */
 215             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 216             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 217
 218             /* LENNARD-JONES DISPERSION/REPULSION */
 219
 220             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 221             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 222             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 223             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 224                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 225             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 226
 227             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 228
 229             /* Update potential sum for this i atom from the interaction with this j atom. */
 230             velec            = _mm256_and_pd(velec,cutoff_mask);
 231             velecsum         = _mm256_add_pd(velecsum,velec);
 232             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 233             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 234
 235             fscal            = _mm256_add_pd(felec,fvdw);
 236
 237             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 238
 239             /* Calculate temporary vectorial force */
 240             tx               = _mm256_mul_pd(fscal,dx00);
 241             ty               = _mm256_mul_pd(fscal,dy00);
 242             tz               = _mm256_mul_pd(fscal,dz00);
 243
 244             /* Update vectorial force */
 245             fix0             = _mm256_add_pd(fix0,tx);
 246             fiy0             = _mm256_add_pd(fiy0,ty);
 247             fiz0             = _mm256_add_pd(fiz0,tz);
 248
 249             fjptrA             = f+j_coord_offsetA;
 250             fjptrB             = f+j_coord_offsetB;
 251             fjptrC             = f+j_coord_offsetC;
 252             fjptrD             = f+j_coord_offsetD;
 253             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 254
 255             }
 256
 257             /* Inner loop uses 54 flops */
 258         }
 259
 260         if(jidx<j_index_end)
 261         {
 262
 263             /* Get j neighbor index, and coordinate index */
 264             jnrlistA         = jjnr[jidx];
 265             jnrlistB         = jjnr[jidx+1];
 266             jnrlistC         = jjnr[jidx+2];
 267             jnrlistD         = jjnr[jidx+3];
 268             /* Sign of each element will be negative for non-real atoms.
 269              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 270              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 271              */
 272             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 273
 274             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 275             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 276             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 277
 278             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 279             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 280             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 281             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 282             j_coord_offsetA  = DIM*jnrA;
 283             j_coord_offsetB  = DIM*jnrB;
 284             j_coord_offsetC  = DIM*jnrC;
 285             j_coord_offsetD  = DIM*jnrD;
 286
 287             /* load j atom coordinates */
 288             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 289                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 290                                                  &jx0,&jy0,&jz0);
 291
 292             /* Calculate displacement vector */
 293             dx00             = _mm256_sub_pd(ix0,jx0);
 294             dy00             = _mm256_sub_pd(iy0,jy0);
 295             dz00             = _mm256_sub_pd(iz0,jz0);
 296
 297             /* Calculate squared distance and things based on it */
 298             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 299
 300             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 301
 302             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 303
 304             /* Load parameters for j particles */
 305             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 306                                                                  charge+jnrC+0,charge+jnrD+0);
 307             vdwjidx0A        = 2*vdwtype[jnrA+0];
 308             vdwjidx0B        = 2*vdwtype[jnrB+0];
 309             vdwjidx0C        = 2*vdwtype[jnrC+0];
 310             vdwjidx0D        = 2*vdwtype[jnrD+0];
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 317             {
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq00             = _mm256_mul_pd(iq0,jq0);
 321             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 322                                             vdwioffsetptr0+vdwjidx0B,
 323                                             vdwioffsetptr0+vdwjidx0C,
 324                                             vdwioffsetptr0+vdwjidx0D,
 325                                             &c6_00,&c12_00);
 326
 327             /* REACTION-FIELD ELECTROSTATICS */
 328             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 329             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 330
 331             /* LENNARD-JONES DISPERSION/REPULSION */
 332
 333             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 334             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 335             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 336             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 337                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 338             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 339
 340             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velec            = _mm256_and_pd(velec,cutoff_mask);
 344             velec            = _mm256_andnot_pd(dummy_mask,velec);
 345             velecsum         = _mm256_add_pd(velecsum,velec);
 346             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 347             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 348             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 349
 350             fscal            = _mm256_add_pd(felec,fvdw);
 351
 352             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 353
 354             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 355
 356             /* Calculate temporary vectorial force */
 357             tx               = _mm256_mul_pd(fscal,dx00);
 358             ty               = _mm256_mul_pd(fscal,dy00);
 359             tz               = _mm256_mul_pd(fscal,dz00);
 360
 361             /* Update vectorial force */
 362             fix0             = _mm256_add_pd(fix0,tx);
 363             fiy0             = _mm256_add_pd(fiy0,ty);
 364             fiz0             = _mm256_add_pd(fiz0,tz);
 365
 366             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 367             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 368             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 369             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 370             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 371
 372             }
 373
 374             /* Inner loop uses 54 flops */
 375         }
 376
 377         /* End of innermost loop */
 378
 379         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 380                                                  f+i_coord_offset,fshift+i_shift_offset);
 381
 382         ggid                        = gid[iidx];
 383         /* Update potential energies */
 384         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 385         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 386
 387         /* Increment number of inner iterations */
 388         inneriter                  += j_index_end - j_index_start;
 389
 390         /* Outer loop uses 9 flops */
 391     }
 392
 393     /* Increment number of outer iterations */
 394     outeriter        += nri;
 395
 396     /* Update outer/inner flops */
 397
 398     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*54);
 399 }
 400 /*
 401  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_256_double
 402  * Electrostatics interaction: ReactionField
 403  * VdW interaction:            LennardJones
 404  * Geometry:                   Particle-Particle
 405  * Calculate force/pot:        Force
 406  */
 407 void
 408 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_256_double
 409                     (t_nblist * gmx_restrict                nlist,
 410                      rvec * gmx_restrict                    xx,
 411                      rvec * gmx_restrict                    ff,
 412                      t_forcerec * gmx_restrict              fr,
 413                      t_mdatoms * gmx_restrict               mdatoms,
 414                      nb_kernel_data_t * gmx_restrict        kernel_data,
 415                      t_nrnb * gmx_restrict                  nrnb)
 416 {
 417     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 418      * just 0 for non-waters.
 419      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 420      * jnr indices corresponding to data put in the four positions in the SIMD register.
 421      */
 422     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 423     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 424     int              jnrA,jnrB,jnrC,jnrD;
 425     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 426     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 427     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 428     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 429     real             rcutoff_scalar;
 430     real             *shiftvec,*fshift,*x,*f;
 431     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 432     real             scratch[4*DIM];
 433     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 434     real *           vdwioffsetptr0;
 435     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 436     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 437     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 438     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 439     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 440     real             *charge;
 441     int              nvdwtype;
 442     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 443     int              *vdwtype;
 444     real             *vdwparam;
 445     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 446     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 447     __m256d          dummy_mask,cutoff_mask;
 448     __m128           tmpmask0,tmpmask1;
 449     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 450     __m256d          one     = _mm256_set1_pd(1.0);
 451     __m256d          two     = _mm256_set1_pd(2.0);
 452     x                = xx[0];
 453     f                = ff[0];
 454
 455     nri              = nlist->nri;
 456     iinr             = nlist->iinr;
 457     jindex           = nlist->jindex;
 458     jjnr             = nlist->jjnr;
 459     shiftidx         = nlist->shift;
 460     gid              = nlist->gid;
 461     shiftvec         = fr->shift_vec[0];
 462     fshift           = fr->fshift[0];
 463     facel            = _mm256_set1_pd(fr->epsfac);
 464     charge           = mdatoms->chargeA;
 465     krf              = _mm256_set1_pd(fr->ic->k_rf);
 466     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 467     crf              = _mm256_set1_pd(fr->ic->c_rf);
 468     nvdwtype         = fr->ntype;
 469     vdwparam         = fr->nbfp;
 470     vdwtype          = mdatoms->typeA;
 471
 472     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 473     rcutoff_scalar   = fr->rcoulomb;
 474     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 475     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 476
 477     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 478     rvdw             = _mm256_set1_pd(fr->rvdw);
 479
 480     /* Avoid stupid compiler warnings */
 481     jnrA = jnrB = jnrC = jnrD = 0;
 482     j_coord_offsetA = 0;
 483     j_coord_offsetB = 0;
 484     j_coord_offsetC = 0;
 485     j_coord_offsetD = 0;
 486
 487     outeriter        = 0;
 488     inneriter        = 0;
 489
 490     for(iidx=0;iidx<4*DIM;iidx++)
 491     {
 492         scratch[iidx] = 0.0;
 493     }
 494
 495     /* Start outer loop over neighborlists */
 496     for(iidx=0; iidx<nri; iidx++)
 497     {
 498         /* Load shift vector for this list */
 499         i_shift_offset   = DIM*shiftidx[iidx];
 500
 501         /* Load limits for loop over neighbors */
 502         j_index_start    = jindex[iidx];
 503         j_index_end      = jindex[iidx+1];
 504
 505         /* Get outer coordinate index */
 506         inr              = iinr[iidx];
 507         i_coord_offset   = DIM*inr;
 508
 509         /* Load i particle coords and add shift vector */
 510         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 511
 512         fix0             = _mm256_setzero_pd();
 513         fiy0             = _mm256_setzero_pd();
 514         fiz0             = _mm256_setzero_pd();
 515
 516         /* Load parameters for i particles */
 517         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 518         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 519
 520         /* Start inner kernel loop */
 521         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 522         {
 523
 524             /* Get j neighbor index, and coordinate index */
 525             jnrA             = jjnr[jidx];
 526             jnrB             = jjnr[jidx+1];
 527             jnrC             = jjnr[jidx+2];
 528             jnrD             = jjnr[jidx+3];
 529             j_coord_offsetA  = DIM*jnrA;
 530             j_coord_offsetB  = DIM*jnrB;
 531             j_coord_offsetC  = DIM*jnrC;
 532             j_coord_offsetD  = DIM*jnrD;
 533
 534             /* load j atom coordinates */
 535             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 536                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 537                                                  &jx0,&jy0,&jz0);
 538
 539             /* Calculate displacement vector */
 540             dx00             = _mm256_sub_pd(ix0,jx0);
 541             dy00             = _mm256_sub_pd(iy0,jy0);
 542             dz00             = _mm256_sub_pd(iz0,jz0);
 543
 544             /* Calculate squared distance and things based on it */
 545             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 546
 547             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 548
 549             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 550
 551             /* Load parameters for j particles */
 552             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 553                                                                  charge+jnrC+0,charge+jnrD+0);
 554             vdwjidx0A        = 2*vdwtype[jnrA+0];
 555             vdwjidx0B        = 2*vdwtype[jnrB+0];
 556             vdwjidx0C        = 2*vdwtype[jnrC+0];
 557             vdwjidx0D        = 2*vdwtype[jnrD+0];
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 564             {
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             qq00             = _mm256_mul_pd(iq0,jq0);
 568             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 569                                             vdwioffsetptr0+vdwjidx0B,
 570                                             vdwioffsetptr0+vdwjidx0C,
 571                                             vdwioffsetptr0+vdwjidx0D,
 572                                             &c6_00,&c12_00);
 573
 574             /* REACTION-FIELD ELECTROSTATICS */
 575             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 576
 577             /* LENNARD-JONES DISPERSION/REPULSION */
 578
 579             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 580             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 581
 582             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 583
 584             fscal            = _mm256_add_pd(felec,fvdw);
 585
 586             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 587
 588             /* Calculate temporary vectorial force */
 589             tx               = _mm256_mul_pd(fscal,dx00);
 590             ty               = _mm256_mul_pd(fscal,dy00);
 591             tz               = _mm256_mul_pd(fscal,dz00);
 592
 593             /* Update vectorial force */
 594             fix0             = _mm256_add_pd(fix0,tx);
 595             fiy0             = _mm256_add_pd(fiy0,ty);
 596             fiz0             = _mm256_add_pd(fiz0,tz);
 597
 598             fjptrA             = f+j_coord_offsetA;
 599             fjptrB             = f+j_coord_offsetB;
 600             fjptrC             = f+j_coord_offsetC;
 601             fjptrD             = f+j_coord_offsetD;
 602             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 603
 604             }
 605
 606             /* Inner loop uses 37 flops */
 607         }
 608
 609         if(jidx<j_index_end)
 610         {
 611
 612             /* Get j neighbor index, and coordinate index */
 613             jnrlistA         = jjnr[jidx];
 614             jnrlistB         = jjnr[jidx+1];
 615             jnrlistC         = jjnr[jidx+2];
 616             jnrlistD         = jjnr[jidx+3];
 617             /* Sign of each element will be negative for non-real atoms.
 618              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 619              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 620              */
 621             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 622
 623             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 624             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 625             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 626
 627             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 628             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 629             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 630             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 631             j_coord_offsetA  = DIM*jnrA;
 632             j_coord_offsetB  = DIM*jnrB;
 633             j_coord_offsetC  = DIM*jnrC;
 634             j_coord_offsetD  = DIM*jnrD;
 635
 636             /* load j atom coordinates */
 637             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 638                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 639                                                  &jx0,&jy0,&jz0);
 640
 641             /* Calculate displacement vector */
 642             dx00             = _mm256_sub_pd(ix0,jx0);
 643             dy00             = _mm256_sub_pd(iy0,jy0);
 644             dz00             = _mm256_sub_pd(iz0,jz0);
 645
 646             /* Calculate squared distance and things based on it */
 647             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 648
 649             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 650
 651             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 652
 653             /* Load parameters for j particles */
 654             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 655                                                                  charge+jnrC+0,charge+jnrD+0);
 656             vdwjidx0A        = 2*vdwtype[jnrA+0];
 657             vdwjidx0B        = 2*vdwtype[jnrB+0];
 658             vdwjidx0C        = 2*vdwtype[jnrC+0];
 659             vdwjidx0D        = 2*vdwtype[jnrD+0];
 660
 661             /**************************
 662              * CALCULATE INTERACTIONS *
 663              **************************/
 664
 665             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 666             {
 667
 668             /* Compute parameters for interactions between i and j atoms */
 669             qq00             = _mm256_mul_pd(iq0,jq0);
 670             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 671                                             vdwioffsetptr0+vdwjidx0B,
 672                                             vdwioffsetptr0+vdwjidx0C,
 673                                             vdwioffsetptr0+vdwjidx0D,
 674                                             &c6_00,&c12_00);
 675
 676             /* REACTION-FIELD ELECTROSTATICS */
 677             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 678
 679             /* LENNARD-JONES DISPERSION/REPULSION */
 680
 681             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 682             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 683
 684             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 685
 686             fscal            = _mm256_add_pd(felec,fvdw);
 687
 688             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 689
 690             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 691
 692             /* Calculate temporary vectorial force */
 693             tx               = _mm256_mul_pd(fscal,dx00);
 694             ty               = _mm256_mul_pd(fscal,dy00);
 695             tz               = _mm256_mul_pd(fscal,dz00);
 696
 697             /* Update vectorial force */
 698             fix0             = _mm256_add_pd(fix0,tx);
 699             fiy0             = _mm256_add_pd(fiy0,ty);
 700             fiz0             = _mm256_add_pd(fiz0,tz);
 701
 702             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 703             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 704             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 705             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 706             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 707
 708             }
 709
 710             /* Inner loop uses 37 flops */
 711         }
 712
 713         /* End of innermost loop */
 714
 715         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 716                                                  f+i_coord_offset,fshift+i_shift_offset);
 717
 718         /* Increment number of inner iterations */
 719         inneriter                  += j_index_end - j_index_start;
 720
 721         /* Outer loop uses 7 flops */
 722     }
 723
 724     /* Increment number of outer iterations */
 725     outeriter        += nri;
 726
 727     /* Update outer/inner flops */
 728
 729     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 730 }